On Mon, 31 Mar 2025 16:03:57 GMT, Per Minborg <pminb...@openjdk.org> wrote:
> Here are the results of the new `MethodHandle` benchmarks: > > ``` > Benchmark Mode Cnt Score Error Units > StableMethodHandleBenchmark.atomic avgt 10 3.676 ? 0.206 ns/op > StableMethodHandleBenchmark.dcl avgt 10 3.766 ? 0.181 ns/op > StableMethodHandleBenchmark.finalMh avgt 10 0.693 ? 0.015 ns/op > StableMethodHandleBenchmark.map avgt 10 7.648 ? 0.308 ns/op > StableMethodHandleBenchmark.nonFinalMh avgt 10 3.381 ? 0.485 ns/op > StableMethodHandleBenchmark.stableMap avgt 10 0.713 ? 0.016 ns/op > StableMethodHandleBenchmark.stableMh avgt 10 0.700 ? 0.030 ns/op > ``` Thanks - I think this shows very clearly what this API is all about ;-) E.g. fast access is one part of the equation, but constant-folding of the accessed result is another very important property. When combined, they allow the kind of performance shown in this benchmark. ------------- PR Comment: https://git.openjdk.org/jdk/pull/23972#issuecomment-2768892425
